直接看代碼例子，有詳細(xì)注釋！！
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
|
import tensorflow as tfimport numpy as npd = np.arange(0,60).reshape([6, 10])# 將array轉(zhuǎn)化為tensordata = tf.data.Dataset.from_tensor_slices(d)# 從data數(shù)據(jù)集中按順序抽取buffer_size個樣本放在buffer中，然后打亂buffer中的樣本# buffer中樣本個數(shù)不足buffer_size，繼續(xù)從data數(shù)據(jù)集中安順序填充至buffer_size，# 此時會再次打亂data = data.shuffle(buffer_size=3)# 每次從buffer中抽取4個樣本data = data.batch(4)# 將data數(shù)據(jù)集重復(fù)，其實就是2個epoch數(shù)據(jù)集data = data.repeat(2)# 構(gòu)造獲取數(shù)據(jù)的迭代器iters = data.make_one_shot_iterator()# 每次從迭代器中獲取一批數(shù)據(jù)batch = iters.get_next()sess = tf.Session()sess.run(batch)# 數(shù)據(jù)集完成遍歷完之后，繼續(xù)抽取的話會報錯：OutOfRangeError |
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
|
In [21]: dOut[21]: array([[ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9], [10, 11, 12, 13, 14, 15, 16, 17, 18, 19], [20, 21, 22, 23, 24, 25, 26, 27, 28, 29], [30, 31, 32, 33, 34, 35, 36, 37, 38, 39], [40, 41, 42, 43, 44, 45, 46, 47, 48, 49], [50, 51, 52, 53, 54, 55, 56, 57, 58, 59]])In [22]: sess.run(batch)Out[22]: array([[ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9], [30, 31, 32, 33, 34, 35, 36, 37, 38, 39], [20, 21, 22, 23, 24, 25, 26, 27, 28, 29], [10, 11, 12, 13, 14, 15, 16, 17, 18, 19]])In [23]: sess.run(batch)Out[23]: array([[40, 41, 42, 43, 44, 45, 46, 47, 48, 49], [50, 51, 52, 53, 54, 55, 56, 57, 58, 59]]) |
從輸出結(jié)果可以看出：
shuffle是按順序?qū)?shù)據(jù)放入buffer里面的；
當(dāng)repeat函數(shù)在shuffle之后的話，是將一個epoch的數(shù)據(jù)集抽取完畢，再進(jìn)行下一個epoch的。
那么，當(dāng)repeat函數(shù)在shuffle之前會怎么樣呢？如下：
|
1
2
3
4
5
|
data = data.repeat(2)data = data.shuffle(buffer_size=3)data = data.batch(4) |
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
|
In [25]: sess.run(batch)Out[25]: array([[10, 11, 12, 13, 14, 15, 16, 17, 18, 19], [20, 21, 22, 23, 24, 25, 26, 27, 28, 29], [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9], [40, 41, 42, 43, 44, 45, 46, 47, 48, 49]])In [26]: sess.run(batch)Out[26]: array([[50, 51, 52, 53, 54, 55, 56, 57, 58, 59], [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9], [30, 31, 32, 33, 34, 35, 36, 37, 38, 39], [30, 31, 32, 33, 34, 35, 36, 37, 38, 39]])In [27]: sess.run(batch)Out[27]: array([[10, 11, 12, 13, 14, 15, 16, 17, 18, 19], [50, 51, 52, 53, 54, 55, 56, 57, 58, 59], [20, 21, 22, 23, 24, 25, 26, 27, 28, 29], [40, 41, 42, 43, 44, 45, 46, 47, 48, 49]]) |
可以看出，其實它就是先將數(shù)據(jù)集復(fù)制一遍，然后把兩個epoch當(dāng)成同一個新的數(shù)據(jù)集，一直shuffle和batch下去。
以上這篇TensorFlow dataset.shuffle、batch、repeat的使用詳解就是小編分享給大家的全部內(nèi)容了，希望能給大家一個參考，也希望大家多多支持服務(wù)器之家。
原文鏈接：https://blog.csdn.net/sgyuanshi/article/details/90183610
